If you don’t know who Gene Amdahl is, you are at the wrong conference, the chairman said when Dr Amdahl – founder of Amdahl Corp and chairman of his latest venture Andor Systems – delivered the keynote address at the third UK Computer Measurement Group annual conference in Harrogate last week. Dr Amdahl gave his views on current developments in the computer industry. These scanned IBM’s strategy from the top-end mainframe market now under attack from Amdahl Corp, to the mid-range where IBM is at loggerheads with DEC, and the schism in the Unix sector where IBM is fighting hard to counter AT&T’s plans. Dr Amdahl also turned the heat up on IBM by announcing that within 18 months his latest venture, Andor Systems Inc, will launch a desktop mainframe with the performance of the IBM 3090-150E. The machine will feature a 64Mb, 24 channels, 10 MIPS performance from a two-chip set on a single board, with fibre optic channels.
For IBM, under increased pressure from IBMulators with the launch by Amdahl Corp of its 5990 mainframe, which has been labelled the 3090 killer, this can only be bad news. Already some industry watchers are saying that IBM’s next generation Summit mainframe is already here – you can get it from Amdahl Corp. So how will IBM respond? Dr Amdahl believes that IBM’s response to the growing competition will be more of the same: making its products more complex and playing its trump card, FUD – spreading Fear, Uncertainty and Doubt. He does not think IBM has any rabbits that it can pull out of its hat. When I was at IBM I was always amazed at how much people thought we had on the shelves. It wasn’t there, Amdahl said, drawing on his experiences at IBM where he spent three years in the early 1950s and from 1960 to 1970. He believes that IBM will delay the launch of Summit until it can come up with something better, because as currently specified Summit will not exceed the performance of the Amdahl 5990. IBM is also going to make it increasingly difficult for IBMulators to keep up with its products by making them unnecessarily complex – Enterprise Systems Architecture with all its additional code was a good example of this strategy. He predicted that this will make it harder for Fujitsu and Hitachi to remain both hardware and software compatible although he believed Amdahl Corp would remain truly compatible. Although he does not believe that the Amdahl 5990 will amount to a 3090 killer Dr Amdahl sees IBM as a company that has lost its direction while aiming to maintain an iron grip on the market. No-one benefits but the company. IBM has put itself in a position were it has lost control of its market place, Amdahl said. IBM is too large. It doesn’t serve anyone else’s interests – customers or competitors – for it to be that large. But this criticism was tempered with All companies operate in their own best interests – its just that some can get away with it – like IBM. Even so, Dr Amdahl found little to praise at IBM. The company’s mid-range strategy was in disarray. Amdahl questioned IBM’s commitment to the Systems 36 and 38, which he reckons are likely to fade into the sunset. As for the other pillar of IBM’s mid-range strategy, IBM deliberately limited the input-output capability of the 9370 so that it did not compete against its 4381 line with the result that the 9370 failed to achieve its goal of meeting DEC head on in the market place. The DEC input-output capability is dismal so it’s not really two giants battling it out, he observed. According to Dr Amdahl and many others too – the DEC VAX architecture is now at the end of its life cycle. And as long as DEC does not unbundle its software as IBM did in 1969 DEC’s advance in the market place will be limited by the inability of hardware makers to make DEC compatible products. The only option for DEC then is for the company to abandon its architecture. Turning to the the IBM 370 range Amdahl said: I don’t think the architecture is a great architecture. The company that has it simply won control over its market place and end users
invested in 370. IBM is the biggest fly catcher of the lot and its victims are said to have customer loyalty. On the Unix front, Dr Amdahl reckons that with the Open Software Foundation, IBM is now trying to prevent AT&T doing what IBM had successfully done in gaining control of its operating environment. AT&T is trying to do what IBM did but IBM doesn’t like it, he said. Not that the good doctor rates Unix. It has gained support in universities because Unix is a good development environment but it is not a good production one. University graduates ask for Unix because Unix’s shortcomings have not yet been brought home to them. So what is the best way of dealing with IBM? I do not aim to antagonise the company or act in a way that will galvanise IBM into responding by putting together 18 great minds and coming up with something we couldn’t. Any company that tries to corner the IBM market will only be swamped by the overwhelming response that will follow as IBM is forced to fight back. I am not going out to pull the rug out from under other companies. But then he is confident that IBM cannot launch a direct challenge against his latest venture. Even if they had our product they would not be motivated to price it low. Their stock would drop and they would make losses. They can’t afford to. A third of IBM’s revenue and half its profits come from mainframe sales. But we will price our product so as not to produce desperation on the part of IBM, he promised. Dr Amdahl concedes that although it was the necessity to hand more and more equity over to Fujitsu Ltd to raise funds that led to his departure from the company, Amdahl Corp – with its 5990 machine can now justifiably argue that the Fujitsu relationship has paid off – it’s the only IBMulator now gaining market share outside of Japan.
On the subject of supercomputers, he believes that IBM’s commitment is to the Vector Facility for the 3090 rather than to its involvement in other supercomputer projects – and he points up the limits to the benefits of vector processing over scalar. His message was that for a work load with between 30% and 70% vector content – the normal range – the performance of the vector processor offers a minor improvement compared with a pure scalar processor, whereas an investment in a scalar processor that is twice as fast doubles performance throughout the normal range. So it may make more sense to invest in a faster scalar processor rather than going to a vector. He also questions the benefits of achieving an overlap by having the scalar process operate while the vector process is operating whenever the computations are independent. In this normal range of 30% to 70% vector there is a rather minor change in performance achieved. It is clearly not worth spending a great deal on hardware to try to achieve such an overlap capacity.